|
Discourse Biases in Anaphor Comprehension Julia Simner*, Alan Garnham**, Martin
Pickering* In passages such as 'John borrowed a jigsaw puzzle. He did it while everyone was out', the 'do it' expression may refer back to the preceding VP ('borrowed the jigsaw puzzle') or to just the NP ('did the jigsaw puzzle'). In three experiments, we examine how the choice of verbal or nominal interpretation is influenced by changes in discourse structure. In this, we illustrate not only the processes that underlie 'do it' comprehension, but those that guide discourse processing more generally. Smyth (1994; Sheldon, 1974) shows that an ambiguous pronoun is preferentially resolved to a preceding NP in a syntactically parallel clause that overlaps in syntactic function. We investigated whether a similar influence plays a role in 'do it' processing, but at the level of discourse comprehension. Across three studies, we manipulated the overlap in discourse structure between antecedent and anaphor expressions. Consider, then, the degree of overlap in 1a-c below. The antecedent context contains a left-edge sentence boundary (#), followed by an NP agent and a VP predicate. Continuations 1a. - c. vary in the extent to which these three features are repeated. In 1a. all three are present, in 1b. the sentence boundary is absent, while in 1c., both the sentence boundary and the immediately preceding NP subject are omitted. 1. # [Sammy]NP [borrowed a jigsaw puzzle]VP Consider now the two types of interpretation afforded by 'do it' in this context. The verbal interpretation ('borrowed the jigsaw puzzle') is a repetition of the event expressed in the antecedent clause, while the nominal interpretation ('did the jigsaw puzzle') introduces a new discourse event. We hypothesised that when the anaphor expression repeats the structure surrounding the antecedent, readers will be more likely to attribute to 'do it' an interpretation that reflects the propositions expressed in the antecedent clause. In other words, as the anaphor context gradually incorporates more structure from the preceding clause (i.e. moving from 1c. to 1b. to 1a.) the number of verbal 'do it' interpretations should increase. Our predictions were supported in an off-line paraphrase task, designed to elicit 'do it' interpretations. We also demonstrate that our hypothesis is superior to alternative accounts based on considerations of causality (e.g. Stevenson et al., 1994), pragmatic biasing, or syntactic parsing (e.g. Smyth, 1994). References |